GoFuckYourself.com - Adult Webmaster Forum - View Single Post

buran · 12-13-2002, 12:33 PM

I've got extensive experience with mysql, postgres, and oracle. I started using msql and migrated to mysql in 1996. Mysql has come a long, long way and is a reasonably solid product. For the majority of web hacks it's going to work great.

The last three software products we've developed have all be designed to work under heavy load with many thousands of requests. We also decided to switch to Postgres to make use of the more advanced features we found ourselves needing.

What you need to do is put a layer inbetween your web requests and the database. You should only serve content directly from the database if there is no other way to do it. What you can do then is to write hits data to flatfiles on the individual machines in your cluster. Then run a cronjob which sucks up that data, collates it, and inserts the reduced data into sql. If you stagger this import on the machines in your cluster, your sql server won't get overburdened and you've no longer got that 1 to 1 association between hits and queries.

Also make sure that you database is optimized, you use the smallest possible column, you have indexes on what you search on and none on the ones you don't, that you set all possible columns as NOT NULL, etc. There are plenty of sites with information on tweaking performance from SQL.

Hope this helps.

12-13-2002, 12:33 PM
buran Confirmed User Join Date: Mar 2002 Location: how'd I get here? Posts: 264	I've got extensive experience with mysql, postgres, and oracle. I started using msql and migrated to mysql in 1996. Mysql has come a long, long way and is a reasonably solid product. For the majority of web hacks it's going to work great. The last three software products we've developed have all be designed to work under heavy load with many thousands of requests. We also decided to switch to Postgres to make use of the more advanced features we found ourselves needing. What you need to do is put a layer inbetween your web requests and the database. You should only serve content directly from the database if there is no other way to do it. What you can do then is to write hits data to flatfiles on the individual machines in your cluster. Then run a cronjob which sucks up that data, collates it, and inserts the reduced data into sql. If you stagger this import on the machines in your cluster, your sql server won't get overburdened and you've no longer got that 1 to 1 association between hits and queries. Also make sure that you database is optimized, you use the smallest possible column, you have indexes on what you search on and none on the ones you don't, that you set all possible columns as NOT NULL, etc. There are plenty of sites with information on tweaking performance from SQL. Hope this helps. __________________ [this signature intentionally left blank]